Detection of fungal infection and Ochratoxin A contamination in stored barley using near-infrared hyperspectral imaging

• Published in: Biosystems engineering Vol. 147; pp. 162 - 173
• Main Authors: Senthilkumar, Thiruppathi, Jayas, Digvir S., White, Noel D.G., Fields, Paul G., Gräfenhan, Tom
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.07.2016
• Subjects: Aspergillus; Classification; Contamination; Fungal infection; Fungi; Glaucus; Hordeum vulgare; Kernels; Near-infrared (NIR) hyperspectral imaging; Ochratoxin A; Penicillium; Penicillium verrucosum; Samples; Statistical analysis; Statistical classifiers; Statistical methods; Wavelengths; Statistical classifiers; Near-infrared (NIR) hyperspectral imaging; Ochratoxin A; Fungal infection
• ISSN: 1537-5110; 1537-5129
• DOI: 10.1016/j.biosystemseng.2016.03.010

Aspergillus glaucus and Penicillium spp. infections and Ochratoxin A contamination were detected in stored barley using a Near-Infrared (NIR) hyperspectral imaging system. Fungal infected samples and Ochratoxin A contaminated samples were subjected to single kernel imaging every two weeks, and acquired three dimensional image data were transformed into two dimensional data. The two dimensional data corresponding to each fungal infected sample and Ochratoxin A contaminated sample were subjected to principal component analysis (PCA) for data reduction, and to identify significant wavelengths. The significant wavelengths 1260, 1310, and 1360 nm corresponding to A. glaucus, Penicillium spp., and non-Ochratoxin A producing Penicillium verrucosum infected kernels and wavelengths 1310, 1360, and 1480 nm corresponding to Ochratoxin A contaminated kernels were obtained based on the highest principal components (PC) factor loadings. Statistical and histogram features from significant wavelengths were extracted and used as input for linear, quadratic, and Mahalanobis statistical classifiers. Pair-wise, two-class, and six-class classification models were developed to differentiate between sterile and infected kernels. The three classifiers differentiated sterile kernels with classification accuracy of more than 94%, fungal infected kernels with more than 80% at initial periods of fungal infection and attained 100% classification accuracy after four weeks of fungal infection. Ochratoxin A contaminated kernels can be differentiated from sterile kernels with a classification accuracy of 100%. Different periods of fungal infection and different levels of Ochratoxin A contamination were discriminated with a classification accuracy of more than 82%. •1260, 1310 and 1360 nm were significant for detecting fungal infection in barley.•1310, 1360 and 1480 nm were significant for detecting Ochratoxin A in barley.•Statistical classification accuracy increased with increase in fungal infection periods.